Fig. 1
From: Visualization of very large high-dimensional data sets as minimum spanning trees
Comparison between TMAP and UMAP on benchmark data sets. Please use the interactive versions of the TMAP visualizations at http://tmap.gdb.tools to see images associated with each point on the map. TMAP explicitly visualizes the relations between as well as within clusters. a, b While UMAP represents the circular nature of the COIL20 subsets, TMAP cuts the circular clusters at the edge of largest difference and joins clusters through an edge of smallest difference. c–f For the MNIST and FMNIST data sets, the tree structure allows for a higher resolution of both variances and errors within clusters as it becomes apparent how sub clusters (branches within clusters) are linked and which true positives connect to false positives. The image data of all three sets was binarized using the average intensity per image as a threshold